J. Brau LCWS 2006 March, 20061 J. Brau LCWS 2006 - Bangalore March, 2006 C. Baltay, W. Emmet, H. Neal, D. Rabinowitz Yale University Jim Brau, O. Igonkina,

J. Brau LCWS 2006 March, 2006

1

J. BrauLCWS 2006 - BangaloreMarch, 2006

C. Baltay, W. Emmet, H. Neal, D. RabinowitzYale University

Jim Brau, O. Igonkina, N. Sinev, D. StromUniversity of Oregon


2

ILC Vertex Detectors

S i D

GLD LDC


3Z= 6.25cm

SiD Vertex Layout

5 barrel layers 4 end disks

SiD00

Design drivers:Smallest radius possibleClear pair background

Seed tracks & vertexingImprove forward regionRole:

Z [cm]

R [cm]

5 Tesla


4

SiD Vertex Detector

• BARREL– 100 sensors– 1750 cm2

• FORWARD– 288 sensors– 2100 cm2


5

ORIGINAL IDEA –Hierarchical array (Macro/Micro) w/SARNOFF


6

Summary• Investigation of Hierarchical Approach

– Macro/Micro Hybrid (50 um ~5 um) Macro only, reduced to 10-15 um pixel

• Completed Macropixel design– 645 transistors– Spice simulation verified design– TSMC 0.18 um -> 40-50 um pixel

• Next phase under consideration– Complete design of Macro pixel– Deliverable –tape out for foundry (this year)

• Future– Fab 50 um Macro pixel design– Then, 10-15 um pixel (Macro pixel)


7


8


9

Background Hits Dominate Vertex Detector • Events of interest are relatively rare –

– less than 1 Hertz. – hit rate in Vertex Detector dominated by background.

• Detailed calculations yield an expected background estimate of

0.03 hits/mm2/Bunch Crossing • However, with considerable uncertainty on this level

of background. – Difficult calculation. – Background will depend on final choice of collider design

details.


10

The Macropixel Array is Critical • Big Pixel size (initially 50 um x 50 um) limits the

tolerance to higher backgrounds. • Therefore important to strive to reduce Big Pixel

size. – Reducing the Big Pixel size to 10 um x 10 um

(or even 15 um x 15 um) makes detector much more tolerant to backgrounds.

– Macropixel Array (Big Pixel size) of 10-15 um might not need complement of micropixels

• simplified design of single layer of "Macropixels" • with time information• Might not need analog information.


11

What Limits the Macropixel Size – Compress Big Pixel size, retaining storage of hit time

information for 4 hits/pixel/bunch-xing – Area needed with present technology (0.25 um?)

• Comparator/counter/latch, etc., circuit • Storage of up to 4 hits, i.e., 14 bits x 4 deep

– Process Technology - how does pixel size scale as process technology goes 0.25 um, 0.13 um, etc?

• What do you need to go to 10 um x 10 um pixels? • Can you estimate the progress of this technology?• What's available today?

– Much more interesting - what will be available - 5 years from now when we need to fabricate the actual devices?;

– How much does it help to reduce max number of time stamps stored to 2 or 3?


12

Readout Procedure and Speed • First, some numbers:

– Consider chips 22 mm x 125 mm = 2750 mm2 – – Total no. of 10 um x 10 um pixels = 27.5 x 1O6 pixels/chip – – Total hits .03 x 2820 x 2750 = 2 x l05 hits/chip/bunch train

• How long does it take to interrogate a pixel to see if it has a hit (presumably look of a single bit flag?)

• How long does it take to read out one hit pixel – X info (up to 2200) - 12 bits + parity = 14 bits – Y info (up to 12500) - 14 bits + parity = 16 bits – Time (up to 3000) - 12 bits + parity = 14 bits

44 bits total • 2 x 105 hits/chip x 44 bits/hit / 50 MHertz = 176 msec• Might divide each chip into parallel readout streams

(10-20) to accommodate higher background rates?


13


14

Charge Spreading • Important to minimize charge spreading

– pixel size sets scale that would reduce need for analog information.

• How small can we keep the charge spreading? – Thickness of expitaxial layer - 10 to 15 um – Possible approach - full depletion of epitaxial

layer • requires high resistivity? - few kohm-cm? or less?

– Depletion voltage, field in epilayer?


15


16

Read Noise • Minimum ionizing particle leaves

88e- /micron in expitaxial layer – 10 um thick epi x 88e-/um = 880 electrons

• GOAL - signal to noise of 10 to 20 – Can we keep read noise below 50 e- or so? – This consideration determines thickness of

the exitaxial layer.


17

Power Consumption

• Keep power to ~100 millwatts/chip (goal) ~4 mW/cm2

• Trade-off noise with power• Make design choices which optimize

noise/power tradeoffs


18

Macropixel Block Diagram

ComparatorDetector

TimingLogic Counter Decoder

14x4Memory

Array

I/OInterface

Write

14

MINIT

RDCLK

ROW_SEL

EmptyY1/Y2

DIO(13:0)

SF_OUT

Vref Bias

RESET

Bias

144


19

Power Dissipation Analysis

• Additional 67- to 100-fold reduction expected by power cycling analog components (0.37 – 0.55 uW)

Detector

Comparator

Timing Logic

Counter/Decoder

Mem. Array

IO Interface

Analog

Digital

ComponentBefore

Optimization

Sub_total

Sub_total

Total

0.05uW

0.07uW

~ 0uW

0.01uW

0.13uW

9.9uW

27.0uW

36.9uW

37.03uW

OptimizedPower Dissipation

11.7uW

35.1uW

46.8uW

Detector

Comparator

Timing Logic

Counter/Decoder

Mem. Array

IO Interface

Analog

Digital

ComponentBefore

Optimization

Sub_total

Sub_total

Total

0.05uW

0.07uW

~ 0uW

0.01uW

0.13uW

9.9uW

27.0uW

36.9uW

37.03uW

OptimizedPower Dissipation

11.7uW

35.1uW

46.8uW


20

Other Considerations• Dark Current

– Keep it small – Sarnoff – will reset array on each bunch

• Should not be a problem • Operating Temperature

– Sarnoff expects modest cooling (<0C adequate)• Device Thickness

– Thinning below 50 um looks feasible• B Field – Lorentz angle


21

Spice Model Verification of Design


22

0.18um1.8V/3.3V

0.13um1.2V/2.5V/3.3V

90nm1.2V/2.5V

65nm1.0V/1.2V/2.5V

40um

30um

20um

15um

Pixel Pitch

Min. Feature Size

10um

45nm0.8V/1.0V/1.2V

2002 2004 2005 2007 2009

Year Available

50um

SARNOFF Response to Question on FutureTechnology Roadmap: Macropixel size estimationvs. Mixed-signal Process Technologies


23

CONCLUSION

• Completed macropixel design– 645 transistors– Spice simulation verifies design– TSMC 0.18 um -> 40-50 um pixel

• Next phase under consideration– Complete design of macro pixel– Deliverable –tape out for foundry

• Future– Fab 50 um pixel chip– Then, 10-15 um pixel


24

EXTRAS


25


26


27

• Power Reduction Ratio = 1/67 to 1/100 (0.552 W to 0.37 W)

• Activate the Detector and the Comparator during the Bunch Train and deactivate rest of the time

200ms 0.95ms

Enable

BunchTrain

2~3ms

Power Reduction Method


28

Documents

J. Brau LCWS 2006 March, 20061 J. Brau LCWS 2006 - Bangalore March, 2006 C. Baltay, W. Emmet, H. Neal, D. Rabinowitz Yale University Jim Brau, O. Igonkina,